When combining multiple independent simulations with `gray -s`, the
statistics in sum-summary.txt are computed by summing the profiles
pointwise and calling `postproc_profiles` with their (supposedly common)
MHD equilibrium.
The subroutine assumes the normalisations of dP/dV⋅dV and J_φ⋅dA are
P_inside(ρ=1), I_cd_inside(ρ=1), however if the individual profiles have
been computed with different versions of Gray, built with different
compilers, or on different machines, the values may be very slightly off.
Further, due to the use of the relation Δρ² = ⟨ρ²⟩ - ⟨ρ⟩², the profile
widths are very sensible to the normalisation, producing wildly
incorrect values.
For example, due to the refactor of `magsurf_data`, a change of about
0.05% in dV resulted in Δρ² being off by 300%.
The file format parsed by read_beam2 also includes the polarisation,
unlike those of read_beam0 and read_beam1.
When running gray standalone, however, we expect the mode to be set by
`antenna.iox` in gray.ini, not by the beam file.
- Avoid logging the same error over and over
- Make all the gray_errors actually warnings
- Replace `large_npl` error with `unstable_beam`, which is actually
the root cause of the former
- Use the gray_main error as exit code
- Replace the `get_free_unit` subroutine with the built-in
`newutin` option of the `open` statement.
- Replace `locatex` with just `locate` + an index offset.
- Replace `inside` with `contour%contains`.
- Merge `vmaxmin` and `vmaxmini` into a single subroutine
with optional arguments.
- Remove unused `range2rect`, `bubble`.
Similarly to eb648039 this change replaces the `equilibrium` module with
a new `gray_equil` module providing the same functionality without using
global variables.
- `read_eqdsk`, `read_equil_an` are replaced by a single `load_equil`
routine that handles all equilibrium kind (analytical, numerical,
and vacuum).
- `scale_equil` is merged into `load_equil`, which besides reading
the equilibrium from file peforms the rescaling and interpolation based
on the `gray_parameters` settings and the equilibrium kind.
To operate on G-EQDSK data specifically, the `change_cocors` and
`scale_eqdsk` are still available. The numeric equilibrium must then
be initialised manually by calling equil%init().
- `set_equil_spline`, `set_equil_an`, `unset_equil_spline`
are completely removed as the module no longer has any internal state.
- `fq` is replaced by `equil%safety`; `bfield` by `equil%b_field`;
`frhotor`, `frhopol` by `equil%pol2tor` and `equil%pol2tor`;
and the remaining subroutines by other methods of `abstract_equil`
retaining the old name.
- the `contours_psi` subroutine is replaced by `equil%flux_contour`,
with a slightly changed invocation but same functionality.
- the `gray_data` type is no longer required ans has been removed: all
the core subroutines now access the input data only though either
`abstract_equil`, `abstract_plasma` or the `limiter` contour.
This change adds a bit of documentation and simplifies the two
(internal) subroutines used to find the horizontal tangent points
and the magnetic O/X point.
Using a closure we can avoid explicitly passing parameters (psi0) to
hybrj1. Previously this required a custom `hybrj1mv` subroutine in
fitpack with an identical interface, except for our extra parameter.
This change replaces the `coreprofiles` module with a new `gray_plasma`
module providing the same functionality without using global variables.
- `read_profiles`, `read_profiles_an` are replaced by a single `load_plasma`
routines that handles both profiles kind (numerical, analytical).
- `scale_profiles` is merged into `load_plasma`, which besides reading
the profiles from file peforms the rescaling and interpolation based
on the `gray_parameters` settings.
- `set_profiles_spline`, `set_profiles_an`, `unset_profiles_spline`
are completely removed as the module no longer has any internal state.
- `density`, `ftemp`, `fzeff` are replaced by the `abstract_plasma`
type which provides the `dens`, `temp` and `zeff` methods for
either `numeric_plasma` or `analytic_plasma` subtypes.
1. Use the `contour` type for limiter and plasma boundary
(rlim, zlim, rbnd, zbnd)
2. Replace `inside` with `contour%contains`
3. Replace `range2rect` with a `contour` interface
4. Remove the limiter module which just re-exports the limiter
as a global; instead just pass the contour object around
This change replaces the output files (Fortran units) with a derived
type called table, that hold the data in memory until further
processing. The data stored in a table can be dumped to a file, as
before, or processed in other ways, for example converted to other
derived type.
This change replaces pointers with automatic arrays to greatly simplify
the memory management in the main subroutine:
- All arrays are defined in a single location and with their final
dimension explicitely shown.
- The allocation/deallocation is performed automatically when
entering/leaving the gray_main routine.
- Remove the Stokes parameters as an intermediate step in the
conversion between Jones vectors and polarisation ellipses.
- Document every single step performed when converting between
different parametrisations and how the polarisation at the
plasma boundary is computed. This includes how everything
was derived from first principles.
- Mark the subroutines as pure.
- Remove `set_pol` entirely.
In situations when multiple beams are traced, either when allowing
multiple plasma crossings (raytracing.ipass > 0) or the initial polarisation
is mixed (raytracing.ipol == .true.), the couplings of all but the first
beam (with least index_rt) were invalid.
The bug is due to the re-use of the psipol,chipol variables as the beams
are traced sequentially over the beam_loop. For the first beam being
traced the psipol,chipol are correctly initialised to the user-defined
value and the resulting coupling is correct.
However, in each subsequent beam the values were not set to those of the
parent beam (or to the user-defined value in the case of the first X
mode beam), but to those of the previous beams (current index_rt - 1).
This change repurposes the psipv,chipv arrays to store the polarisation
of the parent beams, including the initial user-defined value and makes
plasma_in always use these to compute the coupling.
In addition, in the case the polarisation is not immediately known (i.e.
if raytracing.ipol == .false.), this change postpones the computation of
the Jones vector (ext, eyt) from the launch point, if the magnetic
equilibrium is available, to when the ray actually crosses the
plasma boundary.
The original code, besides being strictly incorrect, can lead to
non-negligible alterations to the coupling. This change also mean:
1. most of the functionality of `set_pol` has been merged with
`plasma_in`
2. the polarisation is undefined and the Jones vector is set to the
placeholder value [1, 0] till `plasma_im` is called
Finally, `polarcold` is removed because it's unused.
While technically accepted by GRAY, these indices do not carry a special
meaning, as wrongly implied by the documentation: they are equivalent
to 8, 18 and specifically don't change the meaning of sgnbi,sgni.
Since 24edfdc4 print_prof tabulates the input profiles up to ψ_bnd,
however the torr_curr_psi may fail (producing FPEs) outside ψ=1.
Moreover, torr_curr_psi gives the current on the LFS on the specific
line z=z_maxis, while the flux surface average is usually more
interesting, say when comparing J_φ with Jcd.
This change fixes both issues.
If frhopol is called with a very small ρ_p, less than then step size of
ε_machine^⅓, the two sided finite step differentiation evaluates frhotor
at ρ_p < 0, producing a floating point exception.
In this case we use the single- sided definition.
This adds a proper procedure to rescale and shift a 2D B-spline by
manipulating the coefficients.
Note: the previous code in set_equil_spline did work, but by
transforming the whole partial(i,j) triggered warnings about operations
on uninitialised memory.
When a call to regrid results in ier=-1 (data needs to be refitted with
interpolating spline) the computation of the sum of the residuals can
overflow. In this case we can exit immediately (avoiding raising a FPE)
ignoring the results. The latter need to be recomputed anyway.
The minimisation algorithm may try to evaluate a function outside of its
domain triggering a floating point exception. However, this is not a
concern because it won't affect the final result, so they can be ignored
1. Introduces enumerations (and some booleans) intended to replace all
the magic numbers used throughout the code to represent multiple
choices.
2. Replace the gray_params.sh script a new one that automatically
generates code for all the GRAY parameters by parsing
gray_params.f90.
3. Also generate extra code to accept the enum identifiers as valid
values in the configuration files and command line arguments.
4. Set sensible default values for parameters that are rarely changes.
